lasse klingbeil ipsn 2008 tim wark presenter: shekhar gupta · presenter: shekhar gupta `most of...
TRANSCRIPT
Lasse Klingbeil IPSN 2008Tim Wark
1
Presenter: Shekhar Gupta
Most of the Localization Methods Work for Static NodeUses Range Estimation ( Requires More H/W)
Localization of Mobile Node is ImportantTransport ManagementMilitary Domain Health Services
2
GPS
Cannot work
indoor
Size
Power Consumption Cost
3
Fine Grained:
1. Radio Signal Strength2. Time Difference on Arrival3. Angle of Arrival
Disadvantages:
1. Extensive Computation2. Relatively Higher Hardware Cost3. High Energy4. Size
4
Localization for Human Monitoring in Indoor
Uses Probabilistic Model to Estimate Location
Range Free Algorithm
Combines Data from Gyro Sensor and Previous State to Predict Next State
Uses Map Information to Get More Accurate Estimation
Deployed in Real World
5
Protocol Design for Proposed Algorithm
Step Detection and Heading Measurement
Localization Based on Bayesian Estimation
Simulation and Experimental Results
Conclusion
6
Static NodeAnchor Points in System
Mobile Node1. Nodes Need to be Localized2. Worn by people
8
Sufficient Number of Nodes
Fixed Location
Send data packet to Base Node
Base Node is Connected to Gateway
Gateway Sends Data for Final Processing
9
Always out of
Proximity
Always in the
Proximity
10
Data Preprocessing1. Avoids Congestion2. Saves Energy
Reliable Communication
Uses Data Buffer
Inertial Sensor
11
Localization Algorithm
12
ack
Step Detection1. Data from Accelerometer Sensor2. Heuristic Algorithm for Robustness
Heading (Angle Detection)1. Data from Gyro Sensor2. Integration not Good (Noise)3. Uses Complementary Filter
13
Proximity InformationGiven by Static
Seed Nodes
Improved Mobility Information
Indoor Map InformationGo to Service
Point
Given by Onboard
Inertial Sensors
14
15
Estimate Dynamic System’s State from Noisy Observation
State is Location of Mobile Object
p(xt|xt-1) is Process Model, how the system’s state changing
p(zt|xt) is Measurement Model, likelihood of making observation zt given location xt
Prediction
Correction
16
Initial Prediction
Correction
Estimation
Prediction
Correction
Estimation
18
Works for Non-Gaussian, Non-Linear SystemInformation from Sensor is incorporated for prediction
Map Information is used as Importance factor w[i] for accurate prediction.
Correction Step uses the proximity Information
19
20
Generation of Random PathEvaluation of RMS Error
21
22
Error is Independent of degree of irregularity
Advantage over RSS based Algorithm
Algorithm Deployed in Real Building
23
Without Map With Map
Estimates Location of Mobile Node
Relevant Information Transferred to Central PC
Used Monte Carlo Based Estimation
Evaluated Performance in Real World
RMS Error of 1m
Maximum Error 3.5m
24
Pros:
Bayesian Estimation Gives Pretty Accurate Results.Does not Require Much Hardware.Map Information and Inertial Sensor Data Increased Accuracy.Not Dependent on Correct Radio Propagation Models Like RSS Based Algorithm.
Cons:
If Base Node Dies, Whole System will Fail All Static Nodes are Always Awake, Consumes Lots of Energy
25
Questions ???
26